The Windshield and the Screen
Joanne McNeil is a writer interested in the ways that technology is shaping art, politics, and society.Read full Bio
It was an ordinary, bare-bones ad posted to Craigslist. “Drivers Wanted”. A guy in New York — let’s call him John White — enquired by email. The temp agency phoned him with the address. John had a spotless driving record and he aced the interview. He went through training and they handed him the keys to a Subaru Impreza. The car was white with green doors and branding. John’s daily route was about 40 miles within a 50-mile radius. No dirt roads. He would only drive on clear days. Sometimes the roof rattled in the wind with pressure from above: the sound of a globular multi-lens camera hoisted up with a tripod, mounted to the top of the car. That’s how John became a Google Street View driver.
It wasn’t the most exciting job, but John could play music. He liked Daft Punk and Linkin Park and Muse. Sometimes he listened to radio programs like Car Talk or The Smoking Tire podcast. The domain of his attention stretched beyond the road in front of him. He had to document the world in images too — panoramas, generated and broadcast on a screen beside his steering wheel, appearing at a three second delay. All the images funneled in from all of the camera lenses would be stitched together, forming digital worlds.
After driving an hour, John would park and review the footage on the dashboard. He would touch up the images and delete anything vulgar or obscene. Sometimes if the picture wasn’t right, he would drive through the same streets again. Then he would upload the data and transfer the files to his employer (or drop off hard drives at the office later, if the connection was weak). Six months later, John’s pictures would be added to Google’s maps, for anyone with internet access to see.
A Google Street View car in Los Angeles once captured a picture of Leonard Cohen. It happened a couple of years before he died. He was sitting with an acquaintance on lawn chairs outside his modest home in the Mid-Wilshire neighbourhood. The driver was an accidental paparazzi. Cohen didn’t even notice him.
The picture of Leonard Cohen on Google Street View is part of the database lore circulated on internet forums. Hobbyists and virtual world trawlers trade Street View links—oddities and pranks and wonders, like a man on the street in a horse mask, someone escaping a house with a makeshift rope of bed sheets tied together, and the 'Hallelujah' songwriter in his front yard enjoying a lovely day. A project as ambitious and omnivorous as Street View could only have these wonders, although not on purpose or by design. Street View isn’t photography as aesthetic representation, but the production of leftovers that happen to be images. These images are the husk — the dead skin of a surveillance charade. This archive can be fascinating and even useful to spectators—the users of it. But this data created and cleaned at scale is a source of Google’s power.
Maybe you are John White: driving Google’s unwieldy car around for $15 an hour and uploading a terabyte of data at the end of the day. Or maybe you are Leonard Cohen. You happened to be outside as a driver passed by. Your photograph was taken and your face was blurred. Or maybe you are just a user. But even as a user, you are part of Google’s image machine and harnessed inside its cycle of creating and managing company data.
In 2012, Google put its users to work on the Street View archive. Its reCAPTCHA system, designed to filter bots from the humans visiting websites, began to ask users to identify numbers in pixelated photographs (a process that began a few years earlier, with images of text from Google Books.) These numbers were street addresses from cities digitised by drivers such as John. As Google explains on its product page,
“Every time our CAPTCHAs are solved, that human effort helps digitise text, annotate images, and build machine learning datasets. This in turn helps preserve books, improve maps, and solve hard AI problems”.
The reCAPTCHA security tool moderates website logins for post offices, banks, and government services, which means that even users who don’t have Google accounts can get trapped into performing as Google micro-labourers.
It is likely that you have come across the more recent reCAPTCHA iterations. You click on the checkmark box that says “I’m not a robot”. Then there’s a grid of images and a prompt to select which pictures depict signs or roads or storefronts. Presumably these reCAPTCHA images are part of the machine learning initiative underpinning Waymo, Google’s driverless car program. When a human clicks a picture of a 'storefront', it is classified in terms that algorithms can understand. That’s how the machine learns.
Users have been tapped as distributed teachers to driverless cars. Granted, this is no guarantee of success: the Waymo project keeps missing its deadlines, and even the most ambitious timelines seem to fall beyond the near future. And just as the Street View online archive exists almost as a refuse of data mining, the security function of reCAPTCHA has just about outgrown its usefulness. Bots and spammers crack it all the time. Humans —those “not a robot” — and human labor serves another purpose: a link in the chain, rather than an end in itself.